Your request, rerouted.

One endpoint in front of every scraping API. When a provider gets blocked, the next one runs, and you are told which, and why.

Also an MCP server, for Claude, Codex, Gemini CLI, Cursor and VS Code

No provider works on every target. Teams juggle two or three accounts and the glue code to switch between them. A provider degrades for hours before it fails, and none fails over to a competitor.

And a blocked request looks like a successful one. A captcha page arrives as HTTP 200 with a body. Anything checking status codes records a success and stores the challenge page.

chain

A worked example, in the shape the gateway returns. The first provider answered 200 with a challenge page and the detector named the rule that caught it; the second failed outright; the third served. Your providers charge for all three, so Proxlane reports all three.

Try it in 60 seconds, no provider account

Deploy on Renderfree Deploy on DigitalOcean$5/mo

Runs on your account, not ours. Each deploys a blueprint you can read first.

Bring your own keys Read the source

quickstart

No signup, nothing to paste

One hostname, and the chain is running. The keys already in your environment are found without configuration. This is the gateway: five providers, failover, and the detector, in the first command you type.

terminal

$ npx proxlane doctor

  ok   key:scraperapi     $SCRAPERAPI_KEY set (32 chars)
  ok   key:scrapingbee    $SCRAPINGBEE_KEY set (80 chars)
  ok   cooldowns          on. A provider that just refused a domain is skipped
$ docker run -p 8787:8787 --env-file .env ghcr.io/proxlane/gateway

  proxlane gateway 0.7.0 on :8787
  providers: scraperapi > scrapfly > scrapingbee > brightdata (in order)
  retries:   1 extra at the last provider (PROXLANE_TERMINAL_RETRIES)
$ curl -sD- "localhost:8787/v1?api_key=$KEY&url=https://example.com"

  HTTP/1.1 200 OK
  x-outcome        OK
  x-outcome-class  ok
  x-attempts       2
  x-chain          scraperapi:SOFT_BLOCK>scrapfly:OK
  x-provider-used  scrapfly
  x-cost-estimate  2.000000
  x-cost-unit      provider-credits

$ npx proxlane doctor

  ok   key:scraperapi     $SCRAPERAPI_KEY set (32 chars)
  ok   key:scrapingbee    $SCRAPINGBEE_KEY set (80 chars)
  ok   cooldowns          on. A provider that just refused a domain is skipped
$ docker run -p 8787:8787 --env-file .env ghcr.io/proxlane/gateway

  proxlane gateway 0.7.0 on :8787
  providers: scraperapi > scrapfly > scrapingbee > brightdata (in order)
  retries:   1 extra at the last provider (PROXLANE_TERMINAL_RETRIES)
$ curl -sD- "localhost:8787/v1?api_key=$KEY&url=https://example.com"

  HTTP/1.1 200 OK
  x-outcome        OK
  x-outcome-class  ok
  x-attempts       2
  x-chain          scraperapi:SOFT_BLOCK>scrapfly:OK
  x-provider-used  scrapfly
  x-cost-estimate  2.000000
  x-cost-unit      provider-credits

$ ▊

agents

Give your agent a scraper that admits failure

An MCP server for Claude, Codex, Gemini CLI, Cursor and VS Code. One tool, fetch_page, through the gateway you just started. A block page reaches the model as a failure with a retry hint, never as content to summarise.

terminal

claude mcp add proxlane \
  --env PROXLANE_API_KEY=your-gateway-key \
  -- npx -y @proxlane/sdk@0.1

{ "mcpServers": { "proxlane": {
    "command": "npx", "args": ["-y", "@proxlane/sdk@0.1"],
    "env": { "PROXLANE_API_KEY": "your-gateway-key" } } } }

codex mcp add proxlane \
  --env PROXLANE_API_KEY=your-gateway-key \
  -- npx -y @proxlane/sdk@0.1

gemini mcp add proxlane npx -y @proxlane/sdk@0.1 \
  -e PROXLANE_API_KEY=your-gateway-key

{ "mcpServers": { "proxlane": {
    "command": "npx", "args": ["-y", "@proxlane/sdk@0.1"],
    "env": { "PROXLANE_API_KEY": "your-gateway-key" } } } }

code --add-mcp '{
  "name": "proxlane", "type": "stdio", "command": "npx",
  "args": ["-y", "@proxlane/sdk@0.1"],
  "env": { "PROXLANE_API_KEY": "your-gateway-key" } }'

your-gateway-key is the gateway's PROXLANE_API_KEY from the quickstart. Your provider keys never reach the agent.

fetch_page

you

Summarise the latest changelog at acme.example.

claude code

→ fetch_page url: "https://acme.example/changelog"

gateway

HTTP/1.1 502

x-outcome: HARD_BLOCK
x-outcome-class: blocked
x-attempts: 3

{ "ok": false,
  "outcome": "HARD_BLOCK",
  "class": "blocked",
  "retryable": true }

Every provider was blocked by a bot check on that page, so I have nothing to read and will not guess at what changed. Worth trying again later, not right now.

you

Summarise the latest changelog at acme.example.

claude code

→ fetch_page url: "https://acme.example/changelog"

gateway

HTTP/1.1 502

x-outcome: HARD_BLOCK
x-outcome-class: blocked
x-attempts: 3

{ "ok": false,
  "outcome": "HARD_BLOCK",
  "class": "blocked",
  "retryable": true }

Every provider was blocked by a bot check on that page, so I have nothing to read and will not guess at what changed. Worth trying again later, not right now.

A worked example. The response and tool result are the shapes the gateway and server return; acme.example is a reserved name, not a site.

detection

A 200 is not a success

Detection is table stakes. Showing you the rule that fired is not. Three behaviours instead of the adjective.

The detector reads the body.: A block page returns 200 with a body. The detector calls it SOFT_BLOCK and attaches the rule that fired, so you can see why rather than trust that we looked.
The taxonomy can grow without breaking you.: 21 outcomes, 6 classes. Branch on the class, which never grows; read the outcome for detail. Adding an outcome cannot break your switch.
Failed attempts are still billed, and still reported.: Every attempt is priced, including the ones that failed. A failover that burned two charged hops reports two, not one.

migration

Change one hostname

Same parameters, same status codes. Your code keeps branching on the 404 it already branches on.

curl

curl "https://api.scraperapi.com?api_key=KEY&url=..."
curl "http://localhost:8787/v1?api_key=KEY&url=..."

response

Every answer says how it got there

Branch on the class, read the outcome for detail. Headers follow the chain selected above.

HTTP/1.1 200 OK

x-outcome: OK
x-outcome-class: ok
x-attempts: 3
x-chain: scraperapi:SOFT_BLOCK>scrapingbee:PROVIDER_ERROR>scrapfly:OK
x-provider-used: scrapfly
x-cost-estimate: 3.000000
x-cost-unit: provider-credits
server-timing: gw;dur=2, up;dur=3870, total;dur=3872

pricing

Free on your own keys

You pay your providers directly. We are not in the payment path. The gateway is AGPL and runs on your own machine. There is no account to create and nothing to enter a card into.

pricing

Bring your own keysFree foreverYour provider keys, your provider bill. Proxlane adds nothing to it.
Self-hostFree foreverAGPL-3.0. One container, your infrastructure, no telemetry.
Hosted creditsProvider cost + 5%Not available yet. One bill instead of three, when it lands.

lines

Five scraping APIs, five lines

They are not interchangeable, which is the point. Printed by proxlane providers, straight from the capability registry. One does sessions, one reaches 42 countries, and rendering JavaScript is free on one and 10x the price on another.

proxlane providers

provider	geo	sessions	render-js
scraperapi	all	yes	×10
scrapingbee	42 codes	no	×5
scrapfly	all	no	×6
brightdata	all	no	×1
firecrawl	26 codes	no	×1